Cowboy MCP
Cowboy MCP lets an MCP client securely discover and run WordPress management tools through two gateway tools: cowboy_discover for search/browsing and cowboy_run for execution.
Discover tools by keyword or category, including content, users, media, plugins/themes, WooCommerce, SEO, builders, Wordfence, events, diagnostics, and abilities.
Run WordPress actions such as WP-CLI commands, settings updates, hook inspection, media uploads, content edits, plugin/theme management, database work, and file operations.
Manage site content and structure: posts/pages/CPTs, taxonomies, comments, menus, media, users, and options.
Edit with page builders and integrations: Gutenberg/FSE, ACF, Elementor, Beaver Builder, SiteOrigin, WooCommerce, SEO plugins, forms, caching, Wordfence, and The Events Calendar.
Work safely with safe mode, dry runs, scoped read-only/custom credentials, rate limits, denylists, SSRF protection, and path confinement.
Recover and audit changes using a per-change undo journal, database checkpoints, and an audit log.
Connect via OAuth/browser sign-in or Bearer API keys, with access to resources, resource templates, and workflow prompts.
7 tools for Elementor templates, page content, global styles, and widgets.
Provides 40 tools for WooCommerce management: products, variations, orders, refunds, customers, coupons, tax/shipping/gateway settings, and sales reports.
Core integration for managing WordPress sites, including posts, pages, custom post types, taxonomies, comments, options, users, media, plugins, themes, database queries, WP-CLI, and diagnostics.
Cache provider detection, flush, preload, and settings management for WP Rocket (also supports LiteSpeed and W3TC).
SEO provider detection for Yoast (also supports Rank Math).
Cowboy MCP ๐ค โ Free WordPress MCP Server with Undo
Cowboy MCP is a free, open-source WordPress plugin that turns any WordPress site into a Model Context Protocol (MCP) server over Streamable HTTP, so Claude, ChatGPT, Gemini, Grok, Perplexity, Muse, Cursor, VS Code, Claude Code, Codex, n8n, OpenClaw, Hermes Agent and any other MCP client can manage the site in plain English through 203 tools โ with per-change undo, database checkpoints and an audit log, so it can be trusted on a live site.
Install: WordPress.org plugin directory
Website & guides: cowboymcp.com
Why Cowboy MCP?
Every tool is free. Up to 203 built-in tools plus every ability your plugins register through the WordPress Abilities API โ content, Gutenberg/Site Editor, WooCommerce, users, media, menus, plugins, themes, files, database, WP-CLI, diagnostics, revisions, SEO (Yoast, Rank Math, AIOSEO, SEOPress), The Events Calendar, ACF, Elementor, Beaver Builder, SiteOrigin, Wordfence, caching, forms โ GPL-licensed, no Pro tier, no credits, no usage meter.
Every change is undoable. A per-change undo journal (before-state snapshots, conflict detection, batch undo) plus one-click database checkpoints, with an always-on audit log. Plugin and theme updates take a file backup and a checkpoint first and auto-restore if the post-update health check fails.
Nothing in the middle. The MCP endpoint runs inside your WordPress install. No hosted relay, no account, no telemetry โ your AI client connects straight to your site.
Safe by default. Safe mode (confirmation for destructive tools), dry run on every write tool, per-credential read-only/custom scoping, hashed keys shown once, per-key rate limits, denylists for sensitive options / dangerous SQL / WP-CLI commands, SSRF protection, path confinement to
wp-content, and a Power mode only a human can enable in wp-admin.One-click sign-in for every app. Every app sends you to a sign-in page on your own site: Claude, ChatGPT, Gemini, Grok, Perplexity, Cursor and VS Code with one button or a pasted link, Claude Code, Codex, Gemini CLI and Muse Code with one command, Muse, OpenClaw and Hermes Agent from one pasted message, n8n through OAuth2. The consent screen names the app and lets you grant full access, read-only or a hand-picked list of tools; nothing is granted until you click Approve. API keys remain for scripts, CI and editors without browser sign-in (Windsurf, Cline, Zed).
Works locally too. Local, Studio, MAMP, DevKinsta, wp-env: terminal tools connect with a key as on a live site; Claude Desktop connects through an
mcp-remotebridge the Connections tab generates for you.Context-efficient.
tools/listreturns two gateway tools (cowboy_discover,cowboy_run); the agent discovers and runs the other tools on demand instead of loading 203 schemas into its context. On WordPress 6.9+ every tool is also acowboy-mcp/*ability for WP-CLI, REST and the official MCP Adapter โ with undo.Zero dependencies. Native WordPress APIs only โ no Composer, no npm, no build step, no
wp-admin/includesat request time. Works on hosts without WP-CLI orshell_exec().
"More access than any other MCP offers, easy to use, LOVE the change journal and the checkpoints โ safe if you break something." โ WordPress.org review
Related MCP server: wp-mcp
Tool coverage
Area | Tools | What the agent can do |
Content | 5 posts/pages/CPTs ยท 4 taxonomies ยท 4 comments ยท 4 media ยท 6 menus ยท 1 options | draft, edit, schedule, publish; upload media, fix alt text; build nav menus |
Gutenberg & Site Editor | 15 (8 on classic themes) | read a page as a block tree and edit it by path; block types, patterns, FSE templates/parts, global styles, navigations |
Site administration | 5 users ยท 6 plugins ยท 5 themes ยท 4 files ยท 5 database ยท 3 WP-CLI/system ยท 1 site health | install/update/delete plugins & themes safely, manage roles, edit files in |
Diagnostics | 10 | error log, HTTP & email tests, hooks, transients, REST routes, thumbnails, rewrite rules, snapshot, Connection Doctor |
Safety | 6 rollback ยท 2 batch/audit | list & undo changes, create/list/restore/delete checkpoints, batch execution, audit-log retrieval |
WooCommerce | 40 | products & variations, orders & refunds, customers, coupons, tax/shipping/payment settings, reports |
Wordfence | 17 | scans, blocks, firewall, live traffic, activity, settings |
ACF / Elementor | 9 / 7 | field groups, fields, repeaters / templates, page content, global styles, widgets |
Beaver Builder / SiteOrigin | 7 / 12 | builder pages, layouts with dry-run diff, modules, global settings / layouts and row/widget edits, prebuilt layouts, widgets, settings, Widgets Bundle activation |
Revisions / Events | 3 / 13 | list, diff & restore post revisions / The Events Calendar events, venues, organizers + Events Calendar Pro recurrence |
SEO / Cache / Forms | 4 / 4 / 1 | Yoast, Rank Math, All in One SEO & SEOPress meta read/write/audit / WP Rocket, LiteSpeed, W3TC / WPForms, Gravity Forms, CF7 |
Most other page builders are supported too: a builder that keeps its layout in post content or post meta is edited through the standard post tools, with the same dry run and per-change undo.
Plus 17 read-only resources (incl. wordpress://tools/catalog), 4 resource templates (wordpress://posts/{id}, wordpress://options/{name}, wordpress://plugins/{slug}, wordpress://users/{id}) and 8 workflow prompts with argument auto-completion. Integrations register only when their plugin is active.
Requirements
WordPress 6.2+ (tested up to 7.1)
PHP 8.0+
HTTPS for the OAuth connector and for cloud clients (claude.ai, ChatGPT); plain HTTP is fine for terminal tools on a local site
Installation
Plugins โ Add New, search for Cowboy MCP, install and activate โ or download from WordPress.org.
Settings โ Cowboy MCP โ Generate API Key. Copy it โ it is shown once and stored hashed.
Connect your client (below). The Connection tab shows every snippet pre-filled with your site's endpoint.
Updates arrive through the normal WordPress updates screen.
Connect your app
Every app signs in through a consent screen on your own site; no API key to paste. The full matrix โ plan gates, whether a local site works, and the date each row was last verified end to end โ is at cowboymcp.com/connect.
App | Plan needed | Button or command | Guide |
Claude (claude.ai) | Any Claude plan; Free allows one custom connector. Team and Enterprise: an owner adds it first | Connections tab โ Add to Claude | |
Claude Desktop | Same as claude.ai | Connections tab โ Add to Claude, or Customize โ Connectors โ Add custom connector | |
Claude Code | A Claude plan that includes Claude Code (Pro or Max), or API billing | Connections tab โ Claude Code panel โ run two lines ( | |
ChatGPT | Business, Enterprise or Edu workspace, on the web; Pro connects one server read-only. No Developer mode | Connections tab โ Open ChatGPT, then Plugins โ Add โ Add custom MCP server โ Create as a plugin | |
Codex CLI and app | A ChatGPT plan that includes Codex (Plus, Pro, Business, Enterprise or Edu), or an API key | Connections tab โ Codex panel โ one | |
Perplexity | Pro, Max or Enterprise. Free has no custom connectors | Connections tab โ Open Perplexity, then Plugins (or Customize โ Connectors) โ Add custom โ Add MCP connector | |
Grok | Any, including the free plan | Connections tab โ Open Grok, then grok.com/connectors โ New Connector โ Custom | |
Gemini app | A personal Google account in the US, 18+, Gemini in English; no paid plan listed | Connections tab โ Open Gemini, then Customize โ Connectors โ + Custom | |
Gemini CLI | A Google account; the free tier works | Connections tab โ Gemini CLI panel โ one | |
Muse (Meta) | Free; Meta offers Muse in the United States and Canada | Connections tab โ Muse panel โ Copy message, paste it into a new Muse chat | |
Muse Code | Same Muse account | Connections tab โ Muse Code panel โ Copy message, paste into | |
Cursor | Any Cursor plan | Connections tab โ Add to Cursor | |
VS Code (GitHub Copilot) | Any Copilot plan, including Free | Connections tab โ Install in VS Code | |
n8n | Any n8n Cloud plan, or self-hosted n8n (Community Edition included) | Connections tab โ n8n panel; add an MCP Client Tool node with an MCP OAuth2 credential | |
OpenClaw | A running OpenClaw gateway | Connections tab โ OpenClaw panel โ Copy message, paste into your OpenClaw chat; or | |
Hermes Agent | Hermes Agent by Nous Research; the chat path needs its connection tool, the terminal command works on every install | Connections tab โ Hermes Agent panel โ Copy message, paste into your Hermes chat; or one |
Windsurf, Cline, Zed, LibreChat and any other Streamable HTTP client connect with a scoped API key (config blocks below); Opencode signs in with the same browser flow.
Connecting an agent
The endpoint is https://yoursite.com/wp-json/cowboy-mcp/v1/endpoint (JSON-RPC 2.0 over Streamable HTTP, MCP 2025-06-18).
Claude Code
claude mcp add --transport http your-site https://yoursite.com/wp-json/cowboy-mcp/v1/endpoint \
--header "Authorization: Bearer YOUR_API_KEY"Claude desktop & web, ChatGPT (one-click, no key) โ click Add to Claude / Open ChatGPT on the Connections tab (browser sign-in is switched on for you; for ChatGPT paste the copied link as a custom MCP server), approve the consent screen on your site (choose full, read-only or custom access). Requires a public HTTPS site.
Claude Desktop on a local site โ use the mcp-remote bridge config shown on the Connections tab:
{
"mcpServers": {
"your-site": {
"command": "npx",
"args": ["-y", "mcp-remote", "http://yoursite.local/wp-json/cowboy-mcp/v1/endpoint",
"--header", "Authorization:${AUTH_HEADER}"],
"env": { "AUTH_HEADER": "Bearer YOUR_API_KEY" }
}
}
}Cursor / Windsurf (Devin Desktop) (~/.cursor/mcp.json, ~/.config/devin/mcp_config.json)
{
"mcpServers": {
"your-site": {
"url": "https://yoursite.com/wp-json/cowboy-mcp/v1/endpoint",
"headers": { "Authorization": "Bearer YOUR_API_KEY" }
}
}
}Cline: add "type": "streamableHttp" to the entry. VS Code (.vscode/mcp.json): use servers instead of mcpServers and add "type": "http". Zed: put the same url + headers under context_servers in its settings.
Codex CLI โ Codex reads the key each time it starts, so add the export to your shell profile too.
export COWBOY_MCP_API_KEY="YOUR_API_KEY"
codex mcp add your-site --url https://yoursite.com/wp-json/cowboy-mcp/v1/endpoint --bearer-token-env-var COWBOY_MCP_API_KEYGemini CLI
gemini mcp add --scope user --transport http your-site https://yoursite.com/wp-json/cowboy-mcp/v1/endpoint \
--header "Authorization: Bearer YOUR_API_KEY"Gemini app, Grok, Perplexity, Meta Muse, OpenClaw, Hermes Agent, n8n (no key) โ each has its own panel on the Connections tab: the cloud apps take your connection link as a custom MCP server, Muse, OpenClaw and Hermes take a message you paste into their chat, and n8n's MCP Client Tool node uses MCP OAuth2. All of them sign in on your site through the browser.
Any client that speaks Streamable HTTP with a Bearer header works the same way (n8n, Opencode, LibreChat, your own agent). Opencode: set "oauth": false on the remote server so it uses the key. Step-by-step guides per client: cowboymcp.com.
Quick smoke test with curl
curl -s -X POST https://yoursite.com/wp-json/cowboy-mcp/v1/endpoint \
-H "Authorization: Bearer YOUR_API_KEY" -H "Content-Type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call","params":{"name":"cowboy_run","arguments":{"tool":"wp_site_info","arguments":{}}}}'Safety model
Safe mode (default on): tools annotated
destructiveHintrefuse to run until the call is resent withconfirm: true; the refusal includes a preview.Dry run: every non-read-only tool accepts
dry_run: trueand reports exactly what would change.Undo journal: before-state snapshots for journaled changes (posts, options, users, media, menus, terms, comments, WooCommerce objects, SEO meta, Gutenberg/FSE edits, search-replace rows, plugin/theme packages);
wp_list_changes/wp_undo_change, batch undo, conflict detection, redo-on-undo; 7-day retention by default.Database checkpoints: prefix-scoped dump of the site's tables, atomic restore; up to 5 kept; taken automatically before plugin/theme updates and mutating WP-CLI commands. Checkpoints restore tables, not uploaded files or code.
Audit log: every tool call, error and auth event in
{prefix}cowboy_mcp_audit_log(key, tool, arguments, result, IP); pruned after 30 days; secrets redacted on read.Scoped credentials: API keys and OAuth connections carry
{mode: full|read_only|custom, allowed_tools[]}; enforced at dispatch, also for tools called throughcowboy_runand batches.readOnlyHintis treated as a security boundary.Keys & limits: keys stored as one-way hashes, shown once, revocable individually; per-key rate limit (120/min default); request
Originallowlist; OAuth tokens stored hashed, off by default, admin consent required.Guardrails: protected-option denylist (
siteurl,active_plugins, credentials, the plugin's own settingsโฆ), dangerous-SQL and WP-CLI blocklists (eval,shell,db drop, โฆ) applied on a shell-style tokenizer, SSRF validation on outbound requests,wp-contentpath confinement with atomic writes, a PHP syntax check before every file write and nomu-pluginswrites without Power mode, self-delete and last-administrator protection.Power mode: an admin-only checkbox that lifts the curated guardrails for one-off jobs; it can never be enabled through the API, and it never lifts credential-option protection, secret redaction or self-protection.
Connection Doctor: one-click self-test (HTTPS, reachability, REST, OAuth discovery, host blockers such as Cloudflare challenges, ModSecurity-style WAFs, LiteSpeed caching) with fingerprinted causes and fixes; also
wp cowboy-mcp doctorand thewp_connection_doctortool.
How it compares
Factual, dated comparisons live on the site: all WordPress MCP plugins compared ยท vs Novamira ยท vs AI Engine ยท vs WPVibe ยท vs InstaWP ยท vs the WordPress MCP Adapter ยท self-hosted vs hosted. Short version: the endpoint is self-hosted with no relay or metering, every tool is free, and undo + checkpoints + audit log ship together.
Architecture
cowboy-mcp.php # entry point, constants, activation/uninstall
includes/
class-mcp-transport.php # REST route, JSON-RPC dispatch, sessions (Streamable HTTP)
class-mcp-auth.php # API keys (hashed), Bearer validation, rate limits, Origin allowlist
class-mcp-oauth.php # OAuth 2.1 authorization server (discovery, DCR, consent, tokens)
class-mcp-security.php # denylists, SSRF, SQL/WP-CLI gates, scoping, secret scrubbing
class-mcp-tools.php # registry, gateway meta-tools, dispatch, lazy domain loading
class-mcp-rollback.php # undo journal class-mcp-checkpoint.php # DB checkpoints
class-mcp-audit-log.php # audit table class-mcp-installer.php # WP.org package installer
class-mcp-doctor.php # Connection Doctor class-mcp-compat.php # admin-free reimplementations
class-mcp-resources.php / class-mcp-prompts.php / class-mcp-completion.php
tools/{core,gutenberg,acf,woocommerce,seo,forms,cache,elementor,wordfence}/
admin/ # settings page (Connection, Settings, Activity, Logs, About), assets
languages/ # 12 bundled locales for the admin UI (ru, uk, zh_CN, ja, ko, es, fr, de, pt_BR, it, hi, id)Tool descriptions and error messages returned to agents are intentionally English; the admin UI is translated.
Extensibility
cowboy_mcp_toolsfilter โ register your own tool definitions and handlers.cowboy_mcp_tool_allowedfilter โ block specific tools per request.cowboy_mcp_allowed_originsfilter โ extend the requestOriginallowlist.
Development
npx wp-env start # local WordPress at http://localhost:8890 with this checkout mounted as the plugin
CLI="$(docker container ls -q --filter 'name=^[0-9a-f]+-cli-1$')"
docker exec "$CLI" wp plugin check cowboy-mcp --format=csv # Plugin Check (install it in wp-env first)
find . -name '*.php' -not -path './node_modules/*' -exec php -l {} + # syntax checkNo build step. Pull requests welcome โ keep the WordPress.org review invariants (no wp-admin/includes requires, no path constants, prepared/validated $wpdb queries, escaped output).
Support, reviews, security
Questions and connection problems: the WordPress.org support forum โ paste your Connection Doctor report; topics are usually answered within a day.
Bugs and feature requests: GitHub issues.
Security issues: please report privately through this repository's Security tab (private vulnerability reporting) rather than a public issue.
Reviews: if Cowboy MCP saves you time, a review on WordPress.org helps other site owners find it.
License
GPL-2.0-or-later. ยฉ Andrew Ivanov (februality).
Available Tools
2 toolscowboy_discoverARead-onlyIdempotent
Search or browse available WordPress tools by keyword or category. Returns matching MCP tool definitions with name, description, inputSchema, and annotations.
Examples:
cowboy_discover(query="login failures") โ finds wp_wordfence_update_settings
cowboy_discover(query="error log") โ finds wp_get_php_error_log
cowboy_discover(category="wordfence") โ lists all Wordfence tools
cowboy_discover(query="settings", category="woocommerce") โ WooCommerce settings tools
| Name | Required | Description | Default |
|---|---|---|---|
| query | No | Keyword search, e.g. 'wordfence settings' or 'create post' | |
| category | No | Browse all tools in a category |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already cover the safety profile (readOnly, idempotent, non-destructive, closed-world), and the description adds real value beyond them by disclosing what is returned (name, description, inputSchema, annotations) and that query+category combine as a filter. It does not discuss pagination, result limits, or behavior on zero matches, but the added return-shape detail is substantive.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
One sentence of purpose plus four arrow-form examples, with the purpose front-loaded. Every line earns its place by showing a distinct invocation pattern; there is no filler.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
With no output schema, the description correctly compensates by describing the returned fields. For a zero-required-parameter discovery tool this is nearly complete; a note on result volume or on the discover-then-run workflow would close the remaining gap.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so the baseline is 3, but the description goes further by demonstrating that 'query' and 'category' can be supplied together and are ANDed ('query="settings", category="woocommerce"'), a combining semantic the schema does not state. The enum values themselves are left to the schema.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb+resource ('Search or browse available WordPress tools') and names the two axes of search (keyword or category). It also declares the return payload, which makes it clearly distinct from the sibling cowboy_run, which executes tools rather than finding them.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
The four worked examples make the use cases concrete: keyword-only, category-only, and the combined-query form. What is missing is an explicit statement that this must be called before cowboy_run to obtain a tool name, or any note about when browsing is preferable to searching.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
cowboy_runA
Execute a WordPress tool by name with arguments. Use cowboy_discover first to find the right tool and its inputSchema.
Examples:
cowboy_run(tool="wp_cli", arguments={command: "post list --format=json --posts_per_page=5"})
cowboy_run(tool="wp_wordfence_update_settings", arguments={settings: {maxFailures: 8}})
cowboy_run(tool="wp_get_hooks", arguments={hook: "init"})
| Name | Required | Description | Default |
|---|---|---|---|
| file | No | A file the user attached in the chat (ChatGPT). Only with tool="wp_upload_media"; put title, alt_text, post_id and resize options in arguments. | |
| tool | Yes | Tool name from cowboy_discover results | |
| arguments | No | Tool arguments per the tool's inputSchema |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
Annotations already declare openWorldHint=true and destructiveHint=false, so the description does not need to restate safety. What it omits is meaningful for a generic dispatcher: that side effects are entirely determined by the invoked tool, and that underlying tool errors are surfaced through this call. The examples hint at arbitrary execution but the description never warns that this wrapper can trigger mutating WordPress operations.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
Purpose sentence is front-loaded and short; the example block earns its space by demonstrating the argument shape variance. Three examples is slightly more than needed (two would establish the pattern), but nothing is wasted or buried.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
There is no output schema, so the description should say something about what comes back โ it says nothing about return format, error propagation, or whether the result is the underlying tool's raw output. The nested `file` parameter is fully documented in the schema, so that gap is covered, but the return/error contract is missing for a tool with this much indirection.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema coverage is 100%, so baseline would be 3, but the description adds real value by explaining that `arguments` must conform to the discovered tool's inputSchema and by showing three concrete shapes (flat string command, nested settings object, simple scalar). This makes the indirection between `tool` and `arguments` explicit rather than leaving it to inference.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
States a specific verb and resource ('Execute a WordPress tool by name with arguments') and implicitly separates itself from the only sibling by assigning discovery to cowboy_discover. An agent immediately knows this is the dispatcher, not the catalog.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
'Use cowboy_discover first to find the right tool and its inputSchema' gives an explicit prerequisite and names the alternative tool. It stops short of stating exclusions or failure behavior (e.g., what happens if the tool name is unknown or arguments are invalid), so it is clear context rather than a full when/when-not rule set.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
1 tool update
- Changed
cowboy_run1 field changed- added
Input schema / properties / fileAdded value: +{ + "description": "A file the user attached in the chat (ChatGPT). Only with tool=\"wp_upload_media\"; put title, alt_text, post_id and resize options in arguments.", + "properties": { + "download_url": { + "description": "Temporary download link ChatGPT provides for the file", + "type": "string" + }, + "file_id": { + "description": "ChatGPT file id", + "type": "string" + }, + "file_name": { + "description": "Original file name", + "type": "string" + }, + "mime_type": { + "description": "MIME type, e.g. image/png", + "type": "string" + } + }, + "required": [ + "download_url", + "file_id" + ], + "type": "object" +}
2 tool updates
v1.7.4- First observed
cowboy_discover - First observed
cowboy_run
TDQS
Scored across 2 tools
The two tools have clearly distinct roles: cowboy_discover searches/browses tools, while cowboy_run executes a named tool. There is no overlap, and the descriptions explicitly guide the agent to discover before running.
Both tools follow the exact same cowboy_verb pattern (cowboy_run, cowboy_discover), with no deviations or mixed conventions.
Two tools is slightly below the typical 3-15 range, but it is appropriate for a discovery/execution gateway where the real surface is dynamic. Each tool earns its place and the minimal interface is intentional.
The pair covers the full gateway lifecycle: find a tool (with schema) and execute it. Category browsing and keyword search provide sufficient coverage for accessing the underlying WordPress toolset, with no obvious dead ends.
Maintenance
Related MCP Connectors
Manage WordPress blogs and WooCommerce shops from Claude, ChatGPT, Cursor and other MCP apps.
Free MCP server for WordPress: Claude and ChatGPT connect with OAuth. No account, cap or API key.
Hosted WordPress MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Hosted WooCommerce MCP server for Claude, ChatGPT and Cursor, with each action switched per account.
Related MCP Servers
- -
- AlicenseNot gradedqualityDmaintenanceA Model Context Protocol server that turns WordPress sites into AI-operable surfaces by exposing CRUD primitives for posts, users, comments, and settings. It supports both local STDIO and remote HTTP transports, allowing AI assistants to manage content and site configuration directly.69 npm10MIT
- AlicenseNot gradedqualityCmaintenanceTurns any WordPress site into an MCP server, allowing AI clients to directly control files, database, WP-CLI, PHP, content, and more through declarative abilities without writing code.3GPL 2.0
- AlicenseNot gradedqualityAmaintenanceA free WordPress plugin that turns your site into a governed MCP server, exposing 153 curated WordPress abilities (posts, media, users, WooCommerce, ACF, SEO) as tools for AI agents like Claude and Cursor. Every ability is off by default, scoped to a least-privilege user, capability-gated, and logged.7GPL 2.0